A requirement has arisen in recent years for high-quality reproduction of both printed images and step gradation images (for example an image representing a continuous-tone image as a plurality of step gradation density values) by utilizing a bi-level type of display device such as a gas plasma dot-matrix display which is inherently capable of generating only two levels of display density. This is achieved by generating a spatial gray scale display image. However problems are presented by the prior art methods proposed for implementing such reproduction.
With such a bi-level display device, each picture element generated by the display can be set (e.g. as an element of a dot matrix) in either a light or a dark state. There have been various proposals for implementing pseudo-continuous tone reproduction by such display devices by employing a spatial gray scale, which are based on setting a relatively high proportion of picture elements of the display in the light state within a display region which is to represent a light (i.e. low-density) region of the original image, and a high proportion in the dark state in the case of a dark (i.e. high-density) region of the original image.
The most well-known method of providing such pseudo-continuous tone reproduction is the dither technique, whereby a step gradation representation of a continuous tone image is reproduced on the basis of numbers of dots within each of predetermined areas of the continuous tone image, by utilizing a dither matrix. Threshold values of the dither matrix are compared with the level of an input signal, one picture element at a time, to thereby execute bi-level image conversion processing. However this method has the disadvantage that the step gradation reproduction characteristic and the resolution of the image that is obtained will both depend directly upon the size of the dither matrix, and have a mutually incompatible relationship. Moreover with the dither technique, it is difficult to avoid the generation of moire patterns, particularly when printed images are reproduced.
A method has been proposed in the prior art which is highly effective in overcoming these problems of the dither method. This is the "error diffusion" method, which has been proposed by R. Floyd and L. Steinberg under the title "An Adaptive Algorithm for Spatial Gray Scale", published in the SID 75 Digest, pp 36-37. The basic principles of the proposed method are as follows. To determine whether a picture element of the input signal is to be displayed at the dark or at the light level in the image generated by the bi-level display device, the corresponding level of the input image signal is compared with a threshold value, and a light/dark decision thereby made. The density value (i.e. completely dark or completely light) which is thereby determined for that picture element will in general be in error with respect to the desired density of that element, i.e. with respect to the density of the corresponding picture element in the original image. This error is referred to in the following as the bi-level conversion error. However with the error diffusion method, the effect of this error are substantially reduced by modifying (in accordance with the value of the error) the respective values of desired density to be subsequently used in processing respective ones of a specific set of picture elements which are positioned peripherally adjacent to the object picture element, and have not yet been processed. This modification is performed by apportioning the value of error obtained for that object picture element among these peripheral picture elements, with the error being apportioned in accordance with predetermined fixed factors. Such factors are referred to in the following as apportionment factors.
The term "object picture element" as used herein has the significance of a picture element which is currently being processed, to determine a corresponding bi-level display image value, with data being utilized in that processing which have been obtained beforehand during processing of preceding picture elements.
The error diffusion method is superior to the dither method with regard to image resolution and the reproduction characteristic for a step gradation source image, and enables the degree of generation of moire patterns to be made very small even when a printed image is being reproduced. However in the case of reproduction of an image which has only small amounts of changes in density, such as a computer-generated image having areas of extremely uniform density, the error diffusion method produces regions of texture in the reproduced image. This texture is inherent to the error diffusion method, and for this reason the error diffusion method has not been widely adopted. The reason for generation of this texture is that a fixed relationship is continuously maintained between an object picture element and the aforementioned set of picture elements which are disposed peripherally adjacent to the object picture element, and fixed values are also maintained for the respective proportions by which the bi-level conversion error of the object picture element is apportioned among these peripheral picture elements.
Furthermore, if it is attempted to produce a practical type of digital computation processing circuit for implementing this prior art error diffusion method, it is found that the sum of the aforementioned error apportionment values which are generated during processing of each picture element is not equal to the corresponding bi-level error which is computed for that picture element. This is due to invevitable inaccuracies which arise in the process of computing these error apportionment values, due to discarding of low-significance bits of certain computation results. As a result, the entire bi-level error is not correctly apportioned among the picture elements which are peripherally adjacent to the object picture element, so that all of the possible step gradation density levels of the input signal cannot be properly reproduced in the final image. This is especially true when the input signal level represents a high value or a low value of image density, and hence the range of step gradation image density values which can be reproduced by the prior art error diffusion method is made narrow.
It is a further disadvantage of the prior art is that a plurality of respectively different computations must be executed consecutively in order to process each picture element, so that it is difficult to achieve a high speed of image processing.